lustre/fld/fld_request.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/fld/fld_request.c
  37  *
  38  * FLD (Fids Location Database)
  39  *
  40  * Author: Yury Umanets <umka@clusterfs.com>
  41  */
  42
  43 #ifndef EXPORT_SYMTAB
  44 # define EXPORT_SYMTAB
  45 #endif
  46 #define DEBUG_SUBSYSTEM S_FLD
  47
  48 #ifdef __KERNEL__
  49 # include <libcfs/libcfs.h>
  50 # include <linux/module.h>
  51 # include <linux/jbd.h>
  52 # include <asm/div64.h>
  53 #else /* __KERNEL__ */
  54 # include <liblustre.h>
  55 # include <libcfs/list.h>
  56 #endif
  57
  58 #include <obd.h>
  59 #include <obd_class.h>
  60 #include <lustre_ver.h>
  61 #include <obd_support.h>
  62 #include <lprocfs_status.h>
  63
  64 #include <dt_object.h>
  65 #include <md_object.h>
  66 #include <lustre_req_layout.h>
  67 #include <lustre_fld.h>
  68 #include <lustre_mdc.h>
  69 #include "fld_internal.h"
  70
  71 /* TODO: these 3 functions are copies of flow-control code from mdc_lib.c
  72  * It should be common thing. The same about mdc RPC lock */
  73 static int fld_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
  74 {
  75         int rc;
  76         ENTRY;
  77         spin_lock(&cli->cl_loi_list_lock);
  78         rc = list_empty(&mcw->mcw_entry);
  79         spin_unlock(&cli->cl_loi_list_lock);
  80         RETURN(rc);
  81 };
  82
  83 static void fld_enter_request(struct client_obd *cli)
  84 {
  85         struct mdc_cache_waiter mcw;
  86         struct l_wait_info lwi = { 0 };
  87
  88         spin_lock(&cli->cl_loi_list_lock);
  89         if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
  90                 list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
  91                 cfs_waitq_init(&mcw.mcw_waitq);
  92                 spin_unlock(&cli->cl_loi_list_lock);
  93                 l_wait_event(mcw.mcw_waitq, fld_req_avail(cli, &mcw), &lwi);
  94         } else {
  95                 cli->cl_r_in_flight++;
  96                 spin_unlock(&cli->cl_loi_list_lock);
  97         }
  98 }
  99
 100 static void fld_exit_request(struct client_obd *cli)
 101 {
 102         struct list_head *l, *tmp;
 103         struct mdc_cache_waiter *mcw;
 104
 105         spin_lock(&cli->cl_loi_list_lock);
 106         cli->cl_r_in_flight--;
 107         list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
 108
 109                 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
 110                         /* No free request slots anymore */
 111                         break;
 112                 }
 113
 114                 mcw = list_entry(l, struct mdc_cache_waiter, mcw_entry);
 115                 list_del_init(&mcw->mcw_entry);
 116                 cli->cl_r_in_flight++;
 117                 cfs_waitq_signal(&mcw->mcw_waitq);
 118         }
 119         spin_unlock(&cli->cl_loi_list_lock);
 120 }
 121
 122 static int fld_rrb_hash(struct lu_client_fld *fld,
 123                         seqno_t seq)
 124 {
 125         LASSERT(fld->lcf_count > 0);
 126         return do_div(seq, fld->lcf_count);
 127 }
 128
 129 static struct lu_fld_target *
 130 fld_rrb_scan(struct lu_client_fld *fld, seqno_t seq)
 131 {
 132         struct lu_fld_target *target;
 133         int hash;
 134         ENTRY;
 135
 136         hash = fld_rrb_hash(fld, seq);
 137
 138         list_for_each_entry(target, &fld->lcf_targets, ft_chain) {
 139                 if (target->ft_idx == hash)
 140                         RETURN(target);
 141         }
 142
 143         CERROR("%s: Can't find target by hash %d (seq "LPX64"). "
 144                "Targets (%d):\n", fld->lcf_name, hash, seq,
 145                fld->lcf_count);
 146
 147         list_for_each_entry(target, &fld->lcf_targets, ft_chain) {
 148                 const char *srv_name = target->ft_srv != NULL  ?
 149                         target->ft_srv->lsf_name : "<null>";
 150                 const char *exp_name = target->ft_exp != NULL ?
 151                         (char *)target->ft_exp->exp_obd->obd_uuid.uuid :
 152                         "<null>";
 153
 154                 CERROR("  exp: 0x%p (%s), srv: 0x%p (%s), idx: "LPU64"\n",
 155                        target->ft_exp, exp_name, target->ft_srv,
 156                        srv_name, target->ft_idx);
 157         }
 158
 159         /*
 160          * If target is not found, there is logical error anyway, so here is
 161          * LBUG() to catch this situation.
 162          */
 163         LBUG();
 164         RETURN(NULL);
 165 }
 166
 167 static int fld_dht_hash(struct lu_client_fld *fld,
 168                         seqno_t seq)
 169 {
 170         /* XXX: here should be DHT hash */
 171         return fld_rrb_hash(fld, seq);
 172 }
 173
 174 static struct lu_fld_target *
 175 fld_dht_scan(struct lu_client_fld *fld, seqno_t seq)
 176 {
 177         /* XXX: here should be DHT scan code */
 178         return fld_rrb_scan(fld, seq);
 179 }
 180
 181 struct lu_fld_hash fld_hash[3] = {
 182         {
 183                 .fh_name = "DHT",
 184                 .fh_hash_func = fld_dht_hash,
 185                 .fh_scan_func = fld_dht_scan
 186         },
 187         {
 188                 .fh_name = "RRB",
 189                 .fh_hash_func = fld_rrb_hash,
 190                 .fh_scan_func = fld_rrb_scan
 191         },
 192         {
 193                 0,
 194         }
 195 };
 196
 197 static struct lu_fld_target *
 198 fld_client_get_target(struct lu_client_fld *fld,
 199                       seqno_t seq)
 200 {
 201         struct lu_fld_target *target;
 202         ENTRY;
 203
 204         LASSERT(fld->lcf_hash != NULL);
 205
 206         spin_lock(&fld->lcf_lock);
 207         target = fld->lcf_hash->fh_scan_func(fld, seq);
 208         spin_unlock(&fld->lcf_lock);
 209
 210         if (target != NULL) {
 211                 CDEBUG(D_INFO, "%s: Found target (idx "LPU64
 212                        ") by seq "LPX64"\n", fld->lcf_name,
 213                        target->ft_idx, seq);
 214         }
 215
 216         RETURN(target);
 217 }
 218
 219 /*
 220  * Add export to FLD. This is usually done by CMM and LMV as they are main users
 221  * of FLD module.
 222  */
 223 int fld_client_add_target(struct lu_client_fld *fld,
 224                           struct lu_fld_target *tar)
 225 {
 226         const char *name = fld_target_name(tar);
 227         struct lu_fld_target *target, *tmp;
 228         ENTRY;
 229
 230         LASSERT(tar != NULL);
 231         LASSERT(name != NULL);
 232         LASSERT(tar->ft_srv != NULL || tar->ft_exp != NULL);
 233
 234         if (fld->lcf_flags != LUSTRE_FLD_INIT) {
 235                 CERROR("%s: Attempt to add target %s (idx "LPU64") "
 236                        "on fly - skip it\n", fld->lcf_name, name,
 237                        tar->ft_idx);
 238                 RETURN(0);
 239         } else {
 240                 CDEBUG(D_INFO, "%s: Adding target %s (idx "
 241                        LPU64")\n", fld->lcf_name, name, tar->ft_idx);
 242         }
 243
 244         OBD_ALLOC_PTR(target);
 245         if (target == NULL)
 246                 RETURN(-ENOMEM);
 247
 248         spin_lock(&fld->lcf_lock);
 249         list_for_each_entry(tmp, &fld->lcf_targets, ft_chain) {
 250                 if (tmp->ft_idx == tar->ft_idx) {
 251                         spin_unlock(&fld->lcf_lock);
 252                         OBD_FREE_PTR(target);
 253                         CERROR("Target %s exists in FLD and known as %s:#"LPU64"\n",
 254                                name, fld_target_name(tmp), tmp->ft_idx);
 255                         RETURN(-EEXIST);
 256                 }
 257         }
 258
 259         target->ft_exp = tar->ft_exp;
 260         if (target->ft_exp != NULL)
 261                 class_export_get(target->ft_exp);
 262         target->ft_srv = tar->ft_srv;
 263         target->ft_idx = tar->ft_idx;
 264
 265         list_add_tail(&target->ft_chain,
 266                       &fld->lcf_targets);
 267
 268         fld->lcf_count++;
 269         spin_unlock(&fld->lcf_lock);
 270
 271         RETURN(0);
 272 }
 273 EXPORT_SYMBOL(fld_client_add_target);
 274
 275 /* Remove export from FLD */
 276 int fld_client_del_target(struct lu_client_fld *fld,
 277                           __u64 idx)
 278 {
 279         struct lu_fld_target *target, *tmp;
 280         ENTRY;
 281
 282         spin_lock(&fld->lcf_lock);
 283         list_for_each_entry_safe(target, tmp,
 284                                  &fld->lcf_targets, ft_chain) {
 285                 if (target->ft_idx == idx) {
 286                         fld->lcf_count--;
 287                         list_del(&target->ft_chain);
 288                         spin_unlock(&fld->lcf_lock);
 289
 290                         if (target->ft_exp != NULL)
 291                                 class_export_put(target->ft_exp);
 292
 293                         OBD_FREE_PTR(target);
 294                         RETURN(0);
 295                 }
 296         }
 297         spin_unlock(&fld->lcf_lock);
 298         RETURN(-ENOENT);
 299 }
 300 EXPORT_SYMBOL(fld_client_del_target);
 301
 302 static void fld_client_proc_fini(struct lu_client_fld *fld);
 303
 304 #ifdef LPROCFS
 305 static int fld_client_proc_init(struct lu_client_fld *fld)
 306 {
 307         int rc;
 308         ENTRY;
 309
 310         fld->lcf_proc_dir = lprocfs_register(fld->lcf_name,
 311                                              fld_type_proc_dir,
 312                                              NULL, NULL);
 313
 314         if (IS_ERR(fld->lcf_proc_dir)) {
 315                 CERROR("%s: LProcFS failed in fld-init\n",
 316                        fld->lcf_name);
 317                 rc = PTR_ERR(fld->lcf_proc_dir);
 318                 RETURN(rc);
 319         }
 320
 321         rc = lprocfs_add_vars(fld->lcf_proc_dir,
 322                               fld_client_proc_list, fld);
 323         if (rc) {
 324                 CERROR("%s: Can't init FLD proc, rc %d\n",
 325                        fld->lcf_name, rc);
 326                 GOTO(out_cleanup, rc);
 327         }
 328
 329         RETURN(0);
 330
 331 out_cleanup:
 332         fld_client_proc_fini(fld);
 333         return rc;
 334 }
 335
 336 static void fld_client_proc_fini(struct lu_client_fld *fld)
 337 {
 338         ENTRY;
 339         if (fld->lcf_proc_dir) {
 340                 if (!IS_ERR(fld->lcf_proc_dir))
 341                         lprocfs_remove(&fld->lcf_proc_dir);
 342                 fld->lcf_proc_dir = NULL;
 343         }
 344         EXIT;
 345 }
 346 #else
 347 static int fld_client_proc_init(struct lu_client_fld *fld)
 348 {
 349         return 0;
 350 }
 351
 352 static void fld_client_proc_fini(struct lu_client_fld *fld)
 353 {
 354         return;
 355 }
 356 #endif
 357
 358 static inline int hash_is_sane(int hash)
 359 {
 360         return (hash >= 0 && hash < ARRAY_SIZE(fld_hash));
 361 }
 362
 363 int fld_client_init(struct lu_client_fld *fld,
 364                     const char *prefix, int hash)
 365 {
 366 #ifdef __KERNEL__
 367         int cache_size, cache_threshold;
 368 #endif
 369         int rc;
 370         ENTRY;
 371
 372         LASSERT(fld != NULL);
 373
 374         snprintf(fld->lcf_name, sizeof(fld->lcf_name),
 375                  "cli-%s", prefix);
 376
 377         if (!hash_is_sane(hash)) {
 378                 CERROR("%s: Wrong hash function %#x\n",
 379                        fld->lcf_name, hash);
 380                 RETURN(-EINVAL);
 381         }
 382
 383         fld->lcf_count = 0;
 384         spin_lock_init(&fld->lcf_lock);
 385         fld->lcf_hash = &fld_hash[hash];
 386         fld->lcf_flags = LUSTRE_FLD_INIT;
 387         CFS_INIT_LIST_HEAD(&fld->lcf_targets);
 388
 389 #ifdef __KERNEL__
 390         cache_size = FLD_CLIENT_CACHE_SIZE /
 391                 sizeof(struct fld_cache_entry);
 392
 393         cache_threshold = cache_size *
 394                 FLD_CLIENT_CACHE_THRESHOLD / 100;
 395
 396         fld->lcf_cache = fld_cache_init(fld->lcf_name,
 397                                         FLD_CLIENT_HTABLE_SIZE,
 398                                         cache_size, cache_threshold);
 399         if (IS_ERR(fld->lcf_cache)) {
 400                 rc = PTR_ERR(fld->lcf_cache);
 401                 fld->lcf_cache = NULL;
 402                 GOTO(out, rc);
 403         }
 404 #endif
 405
 406         rc = fld_client_proc_init(fld);
 407         if (rc)
 408                 GOTO(out, rc);
 409         EXIT;
 410 out:
 411         if (rc)
 412                 fld_client_fini(fld);
 413         else
 414                 CDEBUG(D_INFO, "%s: Using \"%s\" hash\n",
 415                        fld->lcf_name, fld->lcf_hash->fh_name);
 416         return rc;
 417 }
 418 EXPORT_SYMBOL(fld_client_init);
 419
 420 void fld_client_fini(struct lu_client_fld *fld)
 421 {
 422         struct lu_fld_target *target, *tmp;
 423         ENTRY;
 424
 425         fld_client_proc_fini(fld);
 426
 427         spin_lock(&fld->lcf_lock);
 428         list_for_each_entry_safe(target, tmp,
 429                                  &fld->lcf_targets, ft_chain) {
 430                 fld->lcf_count--;
 431                 list_del(&target->ft_chain);
 432                 if (target->ft_exp != NULL)
 433                         class_export_put(target->ft_exp);
 434                 OBD_FREE_PTR(target);
 435         }
 436         spin_unlock(&fld->lcf_lock);
 437
 438 #ifdef __KERNEL__
 439         if (fld->lcf_cache != NULL) {
 440                 if (!IS_ERR(fld->lcf_cache))
 441                         fld_cache_fini(fld->lcf_cache);
 442                 fld->lcf_cache = NULL;
 443         }
 444 #endif
 445
 446         EXIT;
 447 }
 448 EXPORT_SYMBOL(fld_client_fini);
 449
 450 static int fld_client_rpc(struct obd_export *exp,
 451                           struct md_fld *mf, __u32 fld_op)
 452 {
 453         struct ptlrpc_request *req;
 454         struct md_fld         *pmf;
 455         __u32                 *op;
 456         int                    rc;
 457         ENTRY;
 458
 459         LASSERT(exp != NULL);
 460
 461         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_FLD_QUERY,
 462                                         LUSTRE_MDS_VERSION, FLD_QUERY);
 463         if (req == NULL)
 464                 RETURN(-ENOMEM);
 465
 466         op = req_capsule_client_get(&req->rq_pill, &RMF_FLD_OPC);
 467         *op = fld_op;
 468
 469         pmf = req_capsule_client_get(&req->rq_pill, &RMF_FLD_MDFLD);
 470         *pmf = *mf;
 471
 472         ptlrpc_request_set_replen(req);
 473         req->rq_request_portal = FLD_REQUEST_PORTAL;
 474         ptlrpc_at_set_req_timeout(req);
 475
 476         if (fld_op != FLD_LOOKUP)
 477                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 478         fld_enter_request(&exp->exp_obd->u.cli);
 479         rc = ptlrpc_queue_wait(req);
 480         fld_exit_request(&exp->exp_obd->u.cli);
 481         if (fld_op != FLD_LOOKUP)
 482                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 483         if (rc)
 484                 GOTO(out_req, rc);
 485
 486         pmf = req_capsule_server_get(&req->rq_pill, &RMF_FLD_MDFLD);
 487         if (pmf == NULL)
 488                 GOTO(out_req, rc = -EFAULT);
 489         *mf = *pmf;
 490         EXIT;
 491 out_req:
 492         ptlrpc_req_finished(req);
 493         return rc;
 494 }
 495
 496 int fld_client_create(struct lu_client_fld *fld,
 497                       seqno_t seq, mdsno_t mds,
 498                       const struct lu_env *env)
 499 {
 500         struct md_fld md_fld = { .mf_seq = seq, .mf_mds = mds };
 501         struct lu_fld_target *target;
 502         int rc;
 503         ENTRY;
 504
 505         fld->lcf_flags |= LUSTRE_FLD_RUN;
 506         target = fld_client_get_target(fld, seq);
 507         LASSERT(target != NULL);
 508
 509         CDEBUG(D_INFO, "%s: Create fld entry (seq: "LPX64"; mds: "
 510                LPU64") on target %s (idx "LPU64")\n", fld->lcf_name,
 511                seq, mds, fld_target_name(target), target->ft_idx);
 512
 513 #ifdef __KERNEL__
 514         if (target->ft_srv != NULL) {
 515                 LASSERT(env != NULL);
 516                 rc = fld_server_create(target->ft_srv, env, seq, mds);
 517         } else {
 518 #endif
 519                 rc = fld_client_rpc(target->ft_exp, &md_fld, FLD_CREATE);
 520 #ifdef __KERNEL__
 521         }
 522 #endif
 523
 524         if (rc == 0) {
 525                 /*
 526                  * Do not return result of calling fld_cache_insert()
 527                  * here. First of all because it may return -EEXISTS. Another
 528                  * reason is that, we do not want to stop proceeding because of
 529                  * cache errors.
 530                  */
 531                 fld_cache_insert(fld->lcf_cache, seq, mds);
 532         } else {
 533                 CERROR("%s: Can't create FLD entry, rc %d\n",
 534                        fld->lcf_name, rc);
 535         }
 536
 537         RETURN(rc);
 538 }
 539 EXPORT_SYMBOL(fld_client_create);
 540
 541 int fld_client_delete(struct lu_client_fld *fld, seqno_t seq,
 542                       const struct lu_env *env)
 543 {
 544         struct md_fld md_fld = { .mf_seq = seq, .mf_mds = 0 };
 545         struct lu_fld_target *target;
 546         int rc;
 547         ENTRY;
 548
 549         fld->lcf_flags |= LUSTRE_FLD_RUN;
 550         fld_cache_delete(fld->lcf_cache, seq);
 551
 552         target = fld_client_get_target(fld, seq);
 553         LASSERT(target != NULL);
 554
 555         CDEBUG(D_INFO, "%s: Delete fld entry (seq: "LPX64") on "
 556                "target %s (idx "LPU64")\n", fld->lcf_name, seq,
 557                fld_target_name(target), target->ft_idx);
 558
 559 #ifdef __KERNEL__
 560         if (target->ft_srv != NULL) {
 561                 LASSERT(env != NULL);
 562                 rc = fld_server_delete(target->ft_srv,
 563                                        env, seq);
 564         } else {
 565 #endif
 566                 rc = fld_client_rpc(target->ft_exp,
 567                                     &md_fld, FLD_DELETE);
 568 #ifdef __KERNEL__
 569         }
 570 #endif
 571
 572         RETURN(rc);
 573 }
 574 EXPORT_SYMBOL(fld_client_delete);
 575
 576 int fld_client_lookup(struct lu_client_fld *fld,
 577                       seqno_t seq, mdsno_t *mds,
 578                       const struct lu_env *env)
 579 {
 580         struct md_fld md_fld = { .mf_seq = seq, .mf_mds = 0 };
 581         struct lu_fld_target *target;
 582         int rc;
 583         ENTRY;
 584
 585         fld->lcf_flags |= LUSTRE_FLD_RUN;
 586
 587         rc = fld_cache_lookup(fld->lcf_cache, seq, mds);
 588         if (rc == 0)
 589                 RETURN(0);
 590
 591         /* Can not find it in the cache */
 592         target = fld_client_get_target(fld, seq);
 593         LASSERT(target != NULL);
 594
 595         CDEBUG(D_INFO, "%s: Lookup fld entry (seq: "LPX64") on "
 596                "target %s (idx "LPU64")\n", fld->lcf_name, seq,
 597                fld_target_name(target), target->ft_idx);
 598
 599 #ifdef __KERNEL__
 600         if (target->ft_srv != NULL) {
 601                 LASSERT(env != NULL);
 602                 rc = fld_server_lookup(target->ft_srv,
 603                                        env, seq, &md_fld.mf_mds);
 604         } else {
 605 #endif
 606                 /*
 607                  * insert the 'inflight' sequence. No need to protect that,
 608                  * we are trying to reduce numbers of RPC but not restrict
 609                  * to them exactly one
 610                  */
 611                 fld_cache_insert_inflight(fld->lcf_cache, seq);
 612                 rc = fld_client_rpc(target->ft_exp,
 613                                     &md_fld, FLD_LOOKUP);
 614 #ifdef __KERNEL__
 615         }
 616 #endif
 617         if (seq < FID_SEQ_START) {
 618                 /*
 619                  * The current solution for IGIF is to bind it to mds0.
 620                  * In the future, this should be fixed once IGIF can be found
 621                  * in FLD.
 622                  */
 623                 md_fld.mf_mds = 0;
 624                 rc = 0;
 625         }
 626
 627         if (rc == 0) {
 628                 *mds = md_fld.mf_mds;
 629
 630                 /*
 631                  * Do not return error here as well. See previous comment in
 632                  * same situation in function fld_client_create().
 633                  */
 634                 fld_cache_insert(fld->lcf_cache, seq, *mds);
 635         } else {
 636                 /* remove 'inflight' seq if it exists */
 637                 fld_cache_delete(fld->lcf_cache, seq);
 638         }
 639         RETURN(rc);
 640 }
 641 EXPORT_SYMBOL(fld_client_lookup);
 642
 643 void fld_client_flush(struct lu_client_fld *fld)
 644 {
 645 #ifdef __KERNEL__
 646         fld_cache_flush(fld->lcf_cache);
 647 #endif
 648 }
 649 EXPORT_SYMBOL(fld_client_flush);